Abstract
Vessel localization is a key component of ultrasound (US) -guided interventional procedures. However, US imaging faces challenges such as insufficient resolution and a low recognition rate in skin vessel detection. In this study, a photoacoustic (PA) -US multi-modal imaging enhancement framework is proposed. High-resolution vascular light absorption maps from photoacoustic microscopy (PAM) provide objective and accurate labels for US images. Accurate enhancement of blood vessels in skin US images is realized by neural networks driven by high-quality data sources. The results of our novel architecture (UIU-Net) on an ex vivo dataset of controllable vascular complexity show that UIU-Net outperforms existing methods in complex vascular morphology. Based on in vivo experiments, UIU-Net predicts vessels with substantial similarity to actual vessels with the best performance compared to conventional methods, with a 25.57% improvement in the similarity coefficient. Extended to the rabbit ear vein puncture scenario, UIU-Net consistently enhances US images of deep microvasculature. This method successfully guided puncture interventions, establishing a US vascular enhancement paradigm guided by PA imaging. It provides an intelligent solution that combines anatomical fidelity with the ability to avoid microvessels, thereby reducing complications in minimally invasive interventional US settings.